A new study from Chinese researchers shows that a simple set of eight clinical and laboratory markers, fed into a logistic regression model, can reliably predict which patients with metabolic dysfunction-associated steatotic liver disease (MASLD) are likely to also have obstructive sleep apnea (OSA). The finding could help clinicians identify high-risk MASLD patients who would benefit from sleep testing, without requiring expensive or time-consuming screening for everyone.
What they found
The multicenter retrospective study, published in BMC Endocrine Disorders, included 510 patients across two Chinese hospitals. The development cohort consisted of 310 patients from Heping Hospital in Changzhi, while an external validation cohort of 200 patients came from Tongling People’s Hospital. All patients had confirmed MASLD and underwent polysomnography, the gold standard test for diagnosing OSA.
The researchers compared seven machine learning models: logistic regression, decision tree, random forest, support vector machine, artificial neural network, XGBoost, and LightGBM. Logistic regression emerged as the best performer, striking the strongest balance between discrimination (how well it separates OSA from non-OSA cases), interpretability (how easily clinicians can understand the model’s reasoning), calibration (how closely predicted probabilities match actual outcomes), and external stability (how well the model generalizes to a new patient population).
The final model uses eight readily available predictors: sex, body mass index (BMI), history of hypertension, type 2 diabetes mellitus (T2DM), alanine aminotransferase (ALT), gamma-glutamyl transferase (GGT), neutrophil-to-lymphocyte ratio (NLR), and the triglyceride-glucose (TyG) index. From these inputs, the model generates an individualized risk score. The researchers translated this into a nomogram, a visual tool that lets clinicians plot a patient’s values and read off their predicted probability of having OSA, without needing to run any software.
Why it matters
MASLD, previously known as non-alcoholic fatty liver disease (NAFLD), affects roughly one in four adults worldwide. It is closely tied to obesity, insulin resistance, and metabolic syndrome. Obstructive sleep apnea, characterized by repeated collapse of the upper airway during sleep, is also extremely common in this population. Studies suggest that 40 to 70 percent of MASLD patients have undiagnosed OSA.
The link between the two conditions runs deeper than shared risk factors. Both diseases are driven by overlapping metabolic and inflammatory mechanisms. Intermittent hypoxia, the repeated drops in blood oxygen that define OSA, triggers oxidative stress and systemic inflammation that can accelerate liver fibrosis in MASLD patients. Conversely, the inflammatory milieu of MASLD may worsen sleep-disordered breathing through its effects on airway tone and central respiratory control. Treating one condition may improve the other, but only if both are diagnosed.
Despite this bidirectional relationship, OSA is dramatically underdiagnosed in MASLD patients. Polysomnography is expensive, inconvenient, and often inaccessible in sleep centers. Clinical prediction tools that can flag which patients need a sleep study could streamline diagnosis, reduce healthcare costs, and improve outcomes for a large and growing patient population.
The study’s findings also highlight a notable pattern in the predictors. Inflammatory markers such as the neutrophil-to-lymphocyte ratio and the TyG index (a proxy for insulin resistance) proved to be strong predictors, reinforcing the central role of metabolic inflammation in linking MASLD and OSA. This suggests that the prediction model is not just a statistical fit but reflects real pathophysiology.
Limitations
The study has important limitations. Its retrospective design means that selection bias and unmeasured confounding cannot be ruled out. All patients came from two hospitals in China, and the model’s performance in other ethnic groups or healthcare settings remains unknown. The study population was predominantly middle-aged and overweight, so generalizability to younger, leaner, or more diverse MASLD patients is uncertain.
The researchers note that future work should test the model prospectively in larger, more diverse cohorts and consider adding biomarkers such as inflammatory cytokines or polysomnography-derived variables that might improve accuracy further. External validation in Western populations with different dietary patterns, genetic backgrounds, and healthcare systems would be particularly valuable.
Bottom line
For clinicians managing MASLD patients, this study offers a practical, data-driven way to gauge OSA risk without adding cost or complexity. The eight predictors are all part of routine clinical care. Sex, BMI, and blood pressure are collected at every visit. ALT, GGT, and fasting glucose (used to calculate the TyG index) are standard lab tests. The neutrophil-to-lymphocyte ratio comes from a complete blood count, which is nearly universal. A nomogram printed on a card or displayed in an electronic medical record could give a clinician a reasonable estimate of OSA probability in under a minute.
For patients, the takeaway is equally direct. If you have MASLD and carry any of the risk factors flagged in this model, especially obesity, hypertension, or diabetes, a sleep evaluation may be worth discussing with your doctor. Untreated OSA does more than disrupt sleep. It worsens liver inflammation, accelerates fibrosis, and raises cardiovascular risk. Catching it early changes the management of both conditions.
For researchers, the study adds to a growing body of evidence that machine learning does not need to be complicated to be useful. In this head-to-head comparison, the simplest model outperformed neural networks and gradient-boosting ensembles, and it did so while remaining transparent enough for a clinician to understand and trust. That combination of simplicity and performance is exactly what clinical adoption requires.
Source
Gao Q, et al. Machine learning-assisted prediction of obstructive sleep apnea in patients with metabolic dysfunction-associated steatotic liver disease: a multicenter study with external validation. BMC Endocrine Disorders. 2026. DOI: 10.1186/s12902-026-02391-y. PMID: 42399746.